Experience, knowledge, and guidelines about intensity modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) have been accumulated over the past decade, but they have largely not been formally modeled to support more efficient and optimal treatment planning and even automate treatment planning in routine cases. In routine clinic practice, IMRT/VMAT treatment planning continues to be a time-consuming iterative process. Each patient presents a unique set of anatomic constraints on how much the dose can be “sculpted” to spare normal tissue, which is currently unknown to the planner. Balancing the competing goalsof target coverage and organs at risk (OARs) sparing is a trial-and-error process guided largely by the planner’s and physician’s personal experience, skill, and knowledge. Further, for any given dose distribution, likely clinical outcomes (e.g. tumor control rate and normal tissue toxicity) are not readily apparent to physicians. Thus, there is a strong need to explicitly organize, model and integrate the available knowledge from various sources into the planning process. We will demonstrate how integration of readily-accessible knowledge into the planning process will improve the efficiency of IMRT/VMAT planning and even automate the planning process completely for certain routine cases. This presents an exciting opportunity to reduce the planning cost while improving the overall quality of resulting plans. We will summarize the state-of-the-art in current approaches for planning knowledge modeling and existing algorithms for automatic generation of best achievable plans. We will also discuss the potentials and challenges to collaboratively extract, represent, integrate, and apply various sources of knowledge in radiation therapy planning in the foreseeable future.